Operator promises to automate various tasks
According to OpenAI CEO Sam Altman, 2025 is set to be a transformative year for AI agents.
In a blog post at the start of the year, Altman predicted a big leap in tools that can automate tasks and independently take actions on behalf of users.
Now, OpenAI has made its first bold attempt to turn that vision into reality.
On January 23, 2025, OpenAI announced a research preview of Operator, a general-purpose AI agent capable of navigating the web and performing tasks autonomously.
Initially available to US users on ChatGPT’s $200 Pro subscription plan, Operator aims to expand to OpenAI’s Plus, Team, and Enterprise tiers in the near future.
Altman shared that Operator will be available in other countries soon although Europe’s rollout may take longer.
The feature is accessible through operator.chatgpt.com, and OpenAI plans to integrate it into all ChatGPT clients eventually.
What Operator Can Do

Operator promises to automate various tasks, from booking travel accommodations to making restaurant reservations and shopping online.
Its interface includes task categories such as shopping, delivery, dining, and travel, each tailored to different types of automation.
When activated, Operator launches a dedicated web browser within a pop-up window, allowing users to see the specific actions it’s taking in real-time.
Users can take over at any moment, as Operator operates through its own browser rather than the user’s primary one.
At the core of Operator is OpenAI’s Computer-Using Agent (CUA) model, which combines the advanced reasoning abilities of OpenAI’s GPT-4o model with computer vision capabilities.
This enables the CUA to interact with websites much like a human would—clicking buttons, navigating menus, and filling out forms—without requiring developer-facing APIs.
Safeguards & Limitations
OpenAI has taken several precautions to ensure Operator operates safely and responsibly.
For example, the CUA model is designed to ask for user confirmation before finalising tasks with external consequences, such as placing an order or sending an email.
OpenAI has also collaborated with companies like DoorDash, eBay, Instacart, Priceline, StubHub, and Uber to ensure Operator respects the terms of service agreements.
However, Operator is not without its limitations.
OpenAI acknowledges that the CUA cannot yet handle complex tasks, such as creating detailed slideshows or managing intricate calendar systems.
Some actions, like banking transactions, require active user supervision, and Operator will not handle sensitive tasks such as inputting credit card details.
Additionally, there are usage rate limits and certain security restrictions. For instance, Operator won’t send emails or delete calendar events in its current iteration.
OpenAI noted:
“Currently, Operator cannot reliably handle many complex or specialised tasks.”
It added that Operator might get “stuck” on complex interfaces or CAPTCHA checks, prompting users to intervene.
Despite these limitations, Operator represents OpenAI’s boldest attempt yet at creating an autonomous AI agent.
Safeguards are in place to detect and pause execution if suspicious activity is detected, and OpenAI employs both automated and human-reviewed pipelines to update these safeguards continuously.
The Broader AI Landscape
The release of Operator comes as other tech giants, including Google and Anthropic, also experiment with AI agents.
These tools promise to go beyond traditional virtual assistants like Siri or Alexa by not only processing information but also taking actions autonomously.
However, this increased functionality also brings safety risks, such as the potential for phishing scams or automated DDoS attacks.
OpenAI’s cautious rollout of Operator reflects its focus on minimising these risks while exploring the technology’s possibilities.
Altman’s ventures extend beyond OpenAI, creating a network of projects that seem to support this AI-centric future.
One such venture is World, a project from Altman and Alex Blania’s Tools for Humanity.
Formerly known as Worldcoin, World aims to address the challenge of distinguishing humans from AI agents online.
By scanning an individual’s eyeball with a silver metal orb, World provides a unique blockchain-based identifier to verify that someone is human.
This “proof of human” tool could also verify AI agents acting on behalf of real people.
Tiago Sada, chief product officer at Tools for Humanity, said:
“This idea of delegating your ‘proof of personhood’ to an agent and letting it act on your behalf is actually super important.”
World’s ID technology could even license AI agents to act on behalf of users, enabling businesses to interact with verified agents without compromising their platforms’ security.
A Future Built Around AI
OpenAI’s partnerships with platforms like Uber and Instacart signal a shift in how businesses interact with users.
As AI agents become more capable, companies may need to adapt to a future where many interactions are facilitated by verified agents rather than individuals.
Tools like World’s “proof of human” system could play a critical role in ensuring trust and security in these interactions.
For Altman, this vision for AI aligns with his other ventures, including Helion Energy, a nuclear fusion startup that could one day power OpenAI’s data centres, and Retro Biosciences, a longevity science startup already leveraging OpenAI’s models.
With Operator’s launch, OpenAI is taking its first major step into the world of autonomous AI agents.
As the technology matures, it could redefine how we use the internet, reshaping industries and unlocking new possibilities.
The year 2025 may very well be the tipping point for AI agents, and OpenAI’s Operator is leading the charge.